The most heavy video ever?

[Delerue]

After searching for a really heavy video to test my new machine, I thought that the 'basketball 60 FPS' (http://www.zshare.net/download/50525013310b465f/) was the winner. But look what I found:

'Crowd Run 2160p.mkv':

General Information
CODEC..........: x264
RUNTIME.......: 0:00:10 (seconds)
SIZE..........: 334 MB
FRAMERATE.....: 50.000
BITRATE.......: 280mbps (CRF22)
RESOLUTION......: 3840x2160
SOURCE..........: scanned 65 mm film

http://rapidshare.com/files/134613382/Crowd.Run.2160p.UHD.CRF22.x264-CtrlHD.rar
http://rapidshare.com/files/134617802/Crowd.Run.2160p.UHD.CRF22.x264-CtrlHD.r00
http://rapidshare.com/files/134621871/Crowd.Run.2160p.UHD.CRF22.x264-CtrlHD.r01
http://rapidshare.com/files/134625808/Crowd.Run.2160p.UHD.CRF22.x264-CtrlHD.r02

1- Doesn't open with CoreAVC 1.85 (last version);
2- Doesn't open with x264 decoder from PWDVD 8 build 2217 (last version from last week);
3- Doesn't open in DXVA with MPC HC (build 849) internal filter;
4- Can be opened with FFDShow (build 2266), but it uses only 2 threads, which is insufficient because even a QX9550 @ 3.6 GHz can't process more than 10 frames (then FFDShow loses sync);
5- Can be opened with MPC HC internal filter (I mean, the one without DXVA), but the framerate is below 10 per second;
6- Can be opened with Divx H.264 decoder (beta 3), but the FPS decreases from the beginning to the end by half, and the best case is worst than MPC internal filter;

Conclusion: http://www.htforum.com/vb/images/more_smilies/bow.gif

[reepa]

CrowdRun is part of a test sequence filmed by SVT (http://svt.se). Here's (http://www.ebu.ch/CMSimages/en/tec_svt_multiformat_v10_tcm6-43174.pdf) a document describing it. The source is scanned 65mm film.

[Dark Shikari]

I was the one who originally sent that guy CrowdRun to encode it for laughs... you should see the original source--48MB per frame!

I did post this here before, but my rip was "only" 120 megabytes. :p

[frogman]

Yes...Great but did you see your cpu / gpu sweat. Mega HD!

[Delerue]

Reepa, thanks for the info. I'll correct.

Dark Shikari, what a coincidence! :)

But the question is: anyone here can decode it in real-time?

[Tagert]

280mpbs, oh my, thats alot :D

[Ajax_Undone]

LOL Higher Hi-Def... That's powerful..

[*.mp4 guy]

Reepa, thanks for the info. I'll correct.

Dark Shikari, what a coincidence! :)

But the question is: anyone here can decode it in real-time?

If you could get coravc (or maybe divx labs decoder) to play it without crashing, and you had an oced c2d with between 4-5ghz clockrate and 1:1 ram, with tight-ish timings, it should be possible. Another thing to consider is your graphics card, at that resolution/framerate, it would be suicidal to use anything but pure hardware accelerated overlay for rendering, VMR9 might not even run in real time without any decoding overhead, at those absurd pixel/s rates; and some gfx cards might not be able to handle scaling and color-space conversion in realtime for such a high resolution (its not that hard, but its not really something that is worried about by gfx companies, they pretty much draw the line at 1080p30)

[Guest]

Why couldn't you upload this to a site that doesn't shut me down after I download one of the files. "You've used too much bandwidth".

Does anyone have a usable link to the stream?

[Ryu77]

Why couldn't you upload this to a site that doesn't shut me down after I download one of the files. "You've used too much bandwidth".

Does anyone have a usable link to the stream?

http://www.mediafire.com/?sharekey=a0977474231d0c7dd2db6fb9a8902bda

There you go. I uploaded it there for you.

[Delerue]

If you could get coravc (or maybe divx labs decoder) to play it without crashing, and you had an oced c2d with between 4-5ghz clockrate and 1:1 ram, with tight-ish timings, it should be possible. Another thing to consider is your graphics card, at that resolution/framerate, it would be suicidal to use anything but pure hardware accelerated overlay for rendering, VMR9 might not even run in real time without any decoding overhead, at those absurd pixel/s rates; and some gfx cards might not be able to handle scaling and color-space conversion in realtime for such a high resolution (its not that hard, but its not really something that is worried about by gfx companies, they pretty much draw the line at 1080p30)

The problem is that CoreAVC doesn't work with this video, neither Cyberlink decoder and any other DXVA mode I know. But we can think: if a QX9550 @ 3.6 GHz plus a 9800 GX2 can only decode it at 10 FPS using 2 threads and MPC internal decoder, so with all 4 threads and the optimizations from CoreAVC team, we can speculate something around 25 FPS, which is only half of the target framerate. Well, let's wait for Nehalem, hehehe.

[squid_80]

But we can think: if a QX9550 @ 3.6 GHz plus a 9800 GX2 can only decode it at 10 FPS using 2 threads and MPC internal decoder, so with all 4 threads and the optimizations from CoreAVC team, we can speculate something around 25 FPS, which is only half of the target framerate.I can tell you you're right on the money here; if CoreAVC could handle this resolution it would benchmark at ~23fps on a stock Q6600. DivX's beta3 decoder does handle it and gives roughly the same.

[Inventive Software]

It is twice the resolution (thus 4 times slower) of 1080p though, so take a 1080p and divide by 4 for a rough estimation of the decoder's performance. And it's mod16 too. ;)

[Delerue]

I can tell you you're right on the money here; if CoreAVC could handle this resolution it would benchmark at ~23fps on a stock Q6600. DivX's beta3 decoder does handle it and gives roughly the same.

I forgot to test with Divx H.264 decoder (beta 3 for now). I tested right now in my machine: C2D @ 3 GHz + 9600 GT + 2 GB DDR2 800 4-5-4-12. The results were not very good. At the beginning, the FPS reach 9, but as long as the video runs the FPS decreases until something around 5. The problem is: if this video had more runtime, probably the FPS would decrease even more. Just to compare, MPC internal filter can decode at constant 8 FPS. Let's hope Divx H.264 decoder final version fix this.

[squid_80]

I don't see that here, I get ~27fps steady for divx, on my stock Q6600 with a low-end 8400GS.

[Sagekilla]

You want something ridiculously difficult? Try downloading one of the scene demos (Metamorphosis, as an example, has an absurdly large amount of motion, especially small scale motion) and running it under kkpature @ 1920x1200 @ 60 or 120 fps (I've done both)

60 fps alone is a bit killer, but since it renders real movement one each frame, 120 fps makes it even harder to compress.

[LoRd_MuldeR]

I captured Metamorphosis at 1680x1050@60fps. CoreAVC is able to play that smoothly on my Q6600.

Higher resolution or framerate would be useless, as it would exceed my screen's resolution or refresh rate...

[Sagekilla]

Indeed it would exceed the refresh rate, but many gamers already do that with people trying to get to unrealistic goals of say, 100 fps.

I for one can capture @ 1920x1200p60 and that tends to bring CoreAVC to its knees on my laptop (Which has only a C2D @ 2.16 GHz)

[Zwitterion]

Indeed it would exceed the refresh rate, but many gamers already do that with people trying to get to unrealistic goals of say, 100 fps.
video game fps != film fps
In movies the interval between two consecutive frames is always the same. In video games this is not always the case. 60 fps could mean 10 frames in the first half of a second and 50 frames in the second (bad example I know).
But take SLI as an example: you could have a nominal frame rate of like 50, but bad synchronisation between the two video cards could lead to a perceived frame rate of 25. Card 1 would show its frame, then card 2 immediately after that, then a big gap, and so on.

[*.mp4 guy]

I didn't realise we had neer linear improvement with core number now (I'm still on a single core). Also can someone please test the difference between a null renderer, overlay, and vmr9, at these resolutions the resuls could be interesting.

[Mug Funky]

hmm... i've got a scanner here that does 35mm in up to 6k super35 (ie about 5600x4200 in academy or thereabouts depending on the camera's gate). i wonder what's the maximum x264 can take?

[LoRd_MuldeR]

hmm... i've got a scanner here that does 35mm in up to 6k super35 (ie about 5600x4200 in academy or thereabouts depending on the camera's gate). i wonder what's the maximum x264 can take?

Can you provide any samples? :D

[Mug Funky]

i'm thinking if there's any film around here that's not owned by someone else...

mainly i'm just teasing :). the arriscan is a sexy machine.

i'm just downloading crowdrun now. it'll be interesting to see what 250D looks like in 65mm up close. i wasn't impressed by their post chain though... simple log-to-lin is fugly as fugly, and shake isn't up to the task IMHO.

[Atamido]

Bah. You want hard to decode, re-encode that to Dirac and try to play it back. I'd be surprised if you could get double digits decoding speed.